Algorithmic welfare. Disaggregating semi-automated fraud detection by spotlighting human judgment in organizational practices

Abstract This paper investigates a risk scoring algorithm used to detect social fraud in the employment sector. It draws on a case study conducted in Austria that spotlights case workers by exploring how the semi-automated fraud detection tool is embedded in, and co-produces, organizational practices in a public social insurance agency. The central question guiding this article is how case workers’ roles, routines, and responsibilities are transformed by the introduction of an algorithm targeting illegal employment, social and wage dumping, and networks of bogus companies. To answer this question, the paper draws on 10 qualitative interviews, short-term ethnographic observations, and a two-day mind-scripting workshop conducted with case workers from different regional social insurance offices. Building on street-level bureaucracy research and critical data studies, the analysis shows how the fraud detection software co-produces new professional identities within welfare organizations, transforms knowledge practices of evidence-gathering, and redistributes responsibility between data science and human judgment. Finally, the paper discusses how semi-automated fraud detection ties into larger trends of welfare states, as part of a shift from care to control.